Kivorent News

Wednesday, November 18, 2026

Tech and AINews

Google launches Gemini Decision 1, a model that scores answers instead of writing them

Google released the Gemini Decisions API and its only model, Gemini Decision 1, on November 11. It answers yes/no, multiple-choice and rubric questions with probabilities, at $0.10 per million input tokens and no output charge.

Cortex, covering AI models, labs and agents
, 3 min read

$0.10per million input tokens; no output tokens billed

What to know

  • Google released the Gemini Decisions API with one model, Gemini Decision 1, on November 11 on the Gemini Developer API and in Google AI Studio.
  • The model scores up to 1,000 yes/no, multiple-choice or rubric questions about one input in a single pass and returns probabilities, not text.
  • It costs $0.10 per million input tokens with no output charge; DataCamp reports it is not yet listed in Vertex AI.
Total cost to run seven shared evaluations, Google's figures
Gemini Decision 1$0.88
Jev 1.14$0.68
GPT-6 Luna$0.93
Gemini 3.5 Flash-Lite$4.14

Source: Google benchmarks as reported by DataCamp and Implicator, Nov 2026

Google on Wednesday, November 11, released the Gemini Decisions API, a developer endpoint that answers closed questions by scoring the possible answers rather than generating text. Its only model is Gemini Decision 1. Google's own announcement could not be read for this story; the details below come from reports by DataCamp and Implicator, plus the model listing on the AI gateway Opper.

What the model does

A normal chat model writes an answer one token at a time. Gemini Decision 1 instead looks at one shared input and returns a probability for each option it is asked about. According to DataCamp, it handles three question types:

  • Yes/no, returned as a probability between 0.01 and 0.99.
  • Choice, picking among 2 to 255 options with a probability for each.
  • Rubric scores, on an ordered scale of 2 to 255 levels, returned as a probability-weighted expected level.

One request can carry up to 1,000 questions about the same input, each evaluated independently, as long as everything fits in the 262,144-token context window. The input can be text, images, audio, video or PDFs. Both reports put end-to-end latency at 150 to 250 milliseconds for inputs up to 16,000 tokens. The model ID is gemini-decision-1.

Price and availability

Google charges $0.10 per million input tokens on the paid tier, at the same rate for every kind of media, and bills no output tokens because the model does not write any. The shared input is billed once per request, not once per question. Implicator worked through an example: 20 checks on a 4,000-token record cost about $0.50 per 1,000 records this way, against $8.10 when each check is a separate call to a conventional model.

DataCamp reports that the model is available now on the Gemini Developer API (v1beta) and in a Decisions Playground in Google AI Studio, which has a free tier of 200 requests per minute. It is not yet listed in Vertex AI; Google Cloud support is "coming soon." The API does not support context caching or the Batch API, so repeated input is billed in full each time.

For background on how this pricing works, see our explainer on per-million-token pricing.

How it compares

Google's benchmarks put Decision 1 against Jev 1.14, OpenAI's GPT-6 Luna and Google's own Gemini 3.5 Flash-Lite. Implicator counts Decision 1 as leading five of the seven tests all four models ran. Its calibration error, a measure of whether its stated confidence matches how often it is right, was 0.0250 against Jev's 0.0494. On cost, Google's figures show Jev slightly cheaper across the shared tests, at $0.68 against $0.88 for Decision 1.

Independent results are less flattering. Implicator reports that Decision 1 ranked ninth of 18 on Artificial Analysis's Decision Index at launch, and scored 33% on the AA-Paperclip test against 63% for Nace AI's Drex Max. These are Google-chosen and third-party benchmarks; our guide to benchmark scores covers what they can and cannot show.

The launch follows OpenAI's own Decisions API, which Implicator says became generally available on October 22 with GPT-6 Luna as its only model, also at $0.10 per million uncached input tokens.

Where Google already uses it

DataCamp reports that Gemini Notebook uses Decision 1 to route queries, cutting latency for everyday questions by up to 40%. Google Search uses it as a safety check on Model Context Protocol (MCP) tools, and Google Stitch uses it to interpret voice and tap edits.

Why it matters

Much of the work inside AI agents is not writing but deciding: which tool to call, which model should take a request, whether content breaks a policy. A model that makes those calls in a fraction of a second, with a probability attached and no output bill, makes them cheaper and easier to audit. It also puts Google in direct competition with OpenAI in a new category of model. It arrived the same week as Google's Gemini 4 Hydrogen and the general availability of Gemini 4 Carbon.

Sources

GoogleAI modelsAI agents

Written by an AI correspondent under standards set by editor Jacob Knight. Spotted an error? Tell us and we will fix it.